In depth Focus: Machine Learning

Definition and Categories:

Machine Learning is based on the concept that algorithms or “models” can:

Machine Learning models can be divided into three broad categories:

Supervised learning: the algorithm must predict the outcome based on the input data and the correct outputs we provide,
Unsupervised learning: there is no output data and the models must identify by patterns and structure within the data,
Reinforcement learning: the algorithm learns the best set of actions to take to achieve a goal.

One of the most commonly used types of machine learning models, where :

A human provides input data and the correct outputs,
The model identifies relationships and dependencies between the input data and the correct output,
To predict the most accurate possible the outcome based on future input data.

Used to solve:

Regression problems to predict continuous values such as price, revenue, sales, satisfaction, and employment income,…
Classification problems where the outcome is discrete class labels or categories such as spam, fraud detection, …

Supervised learning applications:

Less commonly used type of machine learning models, where:

We want to solve problems with little or no idea what our results should look like ⬄ there is no output data,
The models must analyze the data and try to identify hidden patterns and structures based only on the characteristics of the data,

Main unsupervised learning approaches include:

Clustering: one of the most popular unsupervised machine learning techniques used for grouping data points, or objects that are somehow similar.
Association rules: a rule-based method for finding relationships between variables in a given dataset, used for cross-selling, recommendation,…
Dimensionality reduction or and/or feature selection, used to reduce redundant features of a dataset, used to improve data analytics and model performance.

Unsupervised learning applications:

Reinforcement Learning is the science of decision-making.

The program must find the best possible behaviour or path in a specific situation :

By discovering the sequences of actions that maximize the total “reward”,
Through a process of trial-and-error search granting a reward or a penalty at each step of the sequence.

The main difference with unsupervised and supervised machine learning:

Does not rely on a static dataset,
But operates in a dynamic environment and learns from collected experiences.

Complex reinforcement learning problems often rely on deep neural networks, a field known as deep reinforcement learning.

Reinforcement learning applications:

Deep Learning and artificial neural networks can be applied within any of the Machine Learning categories.